Locate the long-history runtime cost

The completed transport screen produced four compatible bounds; both zero-field globe cases reached their unchanged time limit. Preserve those outcomes.

Profile one specified-field, zero-removal generated input and its zero-field, zero-removal counterpart. Use exactly the first 128 packets of each, preserve their measured increments and 0.005-second periods, and query the midpoint of that shortened history. Preserve the original clock limit and sector-4 physical calibration bounds. Construct a new physical-domain source proof for the shortened input; do not transfer a full-history training certificate to a changed source.

These are runtime probes, not position-range, containment or method-usefulness qualifications. Profile the native-enclosure calculation, report time by function and retain any exceptions. Store sanitized function basenames, not private absolute paths. Do not infer full-history speed from profiler timing alone.

Finite allowance: two identities, two propagation bounds and 512 bound packet-steps (two passes through each 128-packet input). One worker with four numerical threads; each bound has a cooperative 45-second limit and the study an overall 120-second limit. Freeze source/input/environment hashes first, charge bounds before execution, and preserve failures without retries or budget expansion. No native trajectory calls, LPs, observed fits, original aircraft reads, protected ESA data or optimizer starts.

Use the profiles to select a specific implementation change. Any replacement must keep conservative bounds and pass independent exact-reference and native controls before another long comparison.